Control System for an All-Terrain Mobile Robot

Transcription

1 Solid State Phenomena Vols (2009) pp Online: (2009) Trans Tech Publications, Switzerland doi: / Control System for an All-Terrain Mobile Robot Paweł Ziemniak 1, a, Dariusz Uciński 1, b and Andreas Paczyński 2, c 1 University of Zielona Góra, ul. Podgórna 50, Zielona Góra, Poland 2 Hochschule Ravensburg-Weingarten, Doggenriedstr., Weingarten, Germany a pawel.ziemniak@hs-weingarten.de, b d.ucinski@issi.uz.zgora.pl, c paczynski@hs-weingarten.de Keywords: robot control, balance, level, inclination, all-terrain, robust Abstract. An idea of a control system architecture for a new wheeled mobile robot is proposed. The robot construction is characterized by an original drive mechanism and constitutes an extension of the previous research performed in Hochschule Ravensburg-Weingarten. In the new construction, the robot is given the ability to rise or lower its chassis. No complicated additional hardware is required as the level of the chassis can be changed by means of torque differences on the wheels. A modular approach is adopted to develop a hierarchical two-level and layered control system. Low and high levels correspond to local and global vehicle control. The low level is described in more detail, defining the layers and providing appropriate justification. Introduction A mobile robot is a combination of various physical (hardware) and computational (software) components [1][2]. In this paper we present the idea of a motion control system for an all-terrain mobile robot which is currently under construction. The robot constitutes a continuation of the previous research held in cooperation between the University of Zielona Góra in Poland and HS Ravensburg-Weingarten in Germany. The previous construction used a unique vehicle steering approach. It had four independent wheels, each supported by exactly one motor. The same motors were used to both drive and steer the whole vehicle by means of torque differences on the wheels. Such an approach reduces the cost since additional hardware for turning is not required and, at the same time, it improves the robustness of the robot by decreasing possible points of failures [3][4]. The proposed new construction benefits from the previous research but extends it by giving the possibility to rise or lower the chassis of the robot. Again it is achieved by means of torque differences, rather than additional complex devices. This paper focuses on the concept of a control system for this new robot and reports some preliminary results. The control system is going to be structured into levels and layers. The levels will distinguish local motion control from global motion control and path planning. In turn, the layers will distinguish particular tasks that should be performed within the levels. Such a modular control system should be easier to implement, test and even to reuse for other vehicles. In the first part of the paper the construction is introduced explaining both mechanical and electrical aspects. Also the possible robot movements are clarified together with an explanation of the way the level of the chassis can be controlled. In the second part the hierarchical control system is presented putting emphasis on the low level control. Then the low level control is subdivided into layers which are discussed in more detail. Finally, some proposals and assumptions are made regarding high level control. Overview of the mobile robot structure The unique mechanical construction consists of a chassis and four wheel-ended legs. There is no mechanical cross-coupling of the legs and thus they can move independently from one another. This kind of construction allows the robot to move and turn in sophisticated ways. Three different kinds of movements can be achieved [4]: All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of Trans Tech Publications, (ID: , Pennsylvania State University, University Park, USA-07/10/15,12:57:43)

2 44 Mechatronic Systems and Materials III Parallel movements moving along a straight line in various directions, Central point rotations rotations in place, turnings in place, Zero-side-slip turnings turnings along any curvature, rotations around a given point. The legs will turn as a result of torque differences on the wheels and the inertia of the robot. In order to make the control possible, the axles have to be equipped with encoders that measure the current angles of the legs. The legs are also designed in the way that allows rising or lowering the whole robot chassis or just some of its corners. It is achieved by shifting the wheel out of the vertical axis (see Fig. 1). Changes in the shift (s) imply changes in the angle α and, consequently, the level of chassis (h). h α s Fig. 1 Rising and lowering the chassis In order to better understand the concept, Fig. 2 presents an example of driving downhill. The rear legs are maximally shifted thus it is the maximum slope that can be driven with desired platform level. Similar applies to driving uphill. The large shift will make turning difficult as the wheels are required to move on a larger circle. Fig. 2 Driving downhill An additional position encoder is required for the horizontal axis to measure the angle α, from which the level of chassis can be derived. Given all angles, the level of the chassis can be determined but only when the terrain is also known. As the robot is supposed to be an all-terrain vehicle, it must be made aware of the terrain it moves on. To provide necessary data, a 2D inclination sensor is mounted at the centre of chassis.

3 Solid State Phenomena Vols Fig. 3 Model of all-terrain mobile robot Figure 3 presents the model of the vehicle. The prototype will be built from four almost identical units. Each unit will consist of a wheel driven by one motor mounted on the lower part of the leg, a motor controller, position encoders and brakes in both vertical and horizontal axles. The only difference between units is that one will be supplied by a more advanced, programmable version of the motor controller. All devices are going to be connected using a CAN bus. The bus will also connect all four units, the inclination sensor, an additional computer for higher level control (environment analysis, path planning) and any further devices. The lower level control will be realized within the programmable motor controller. Proposed hierarchical control system One of software engineering principles suggests developing software using modular constructions. The same, with even more importance, applies to robot design [5]. It will be quite natural to divide the control system into two main parts: low level and high level control. The former is the one closest to the hardware. Very often it has to work in real-time with minimal delays. As a good example, the motor speed controller can be given. It has to continuously adjust the current in order to maintain a constant speed of the motor. A correct decision usually has to be made in time less than several milliseconds. By contrast, for high level control the times can be even up to seconds, e.g., as far as path planning is concerned. Also the cost of wrong decisions or the lack of decisions at all might be dependant on the control level. Applying high currents for a long time might damage not only the motor itself but also the mechanics when the maximum permissible speed is exceeded. For higher level control the cost strictly depends on the environment. Choosing a non-optimal path is not critical unless battery power or fuel is limited. Hitting an obstacle might not be yet that dangerous, but falling from the stairs or getting into water when the device is not water proof might lead to a catastrophe. Subdivision into low and high control levels has some additional advantages. The higher level abstracts away from hardware and thus it can be made more universal. Separated levels (software modules) are easier to develop, test and debug. Going further, the subdivision can be made within those two levels of the control system. Subdividing them into additional layers provides even more advantages than was mentioned for the main division. Figure 4 displays the proposed architecture of the control system with the lower level subdivided into layers. The higher level will be supported by additional hardware that will provide more information about the environment.

4 46 Mechatronic Systems and Materials III High level control Low level control Driving control Chassis level control Balance control Motor position/speed control Low level control system Motors, brakes, encoders, inclinometer Fig. 4 Control system hierarchy Additional sensors: cameras, proximity sensors, gps The sequential arrangement of layers provides the ability to analyze or modify commands passed among them. The analysis can help predicting what might happen to the robot after issuing a command, e.g., when taking a turn at an excessive speed the robot might fall. The ability to modify commands or even to refuse them at all can increase the overall system robustness. Another advantage of structuring the control system is the ease of implementation of the simulation engine. As proposed in [6], the physical robot might be replaced by a suitable robot daemon a software layer that simulates the robot. To this end, only the lowest software layer requires the modification to be able to act with both real hardware and the simulation model. Of course, first the simulation model has to be built. The proposed layers correspond to the main tasks of the low level control system. There are several reasons for the division and these will be discussed separately for each layer. Motor position/speed control. It is the lowest possible control as all other layers will strictly depend on it. Here the demanded acceleration and velocity of the wheel will be assured. The appropriate software (PID controller) is already available in the motor controllers. Balance control. The layer will be responsible for maintaining the balance of the robot. The control is crucial because just a short and sharp turn taken with an excessive speed may make the robot fall. To avoid this situation, balance control must take precedence over other activities. The same situation happens with a human body. When balance is in danger, a quick recovery action is taken. We are not fully aware of what we are doing as the action is unconscious and seems to be automatic. We call it the reflex action. A similar mechanism can be applied to the robot. The balance control works all the time by checking the inclination. When it exceeds some predefined value, a recovery action is taken. In order to increase robustness, an analysis of commands sent from upper layers can be made using appropriate models. The model response would indicate what action should be taken in order to handle balance in a better manner. Such control can be referred to as model predictive control and relies on determination of system inputs by means of model outputs [7]. When accelerating (or decelerating) while moving forward (or backward), it is possible to estimate appropriate forces and to lower or rise the appropriate parts of the chassis in advance to keep the balance in a better manner. The main variables that need to be included in the model are the current and desired velocities, accelerations and positions in horizontal axles of all the wheels, current inclination, friction forces, etc. A slightly more complicated situation occurs while turning as a more precise model is required. Additional variable have to be taken into account, e.g., the centrifugal force and angular positions of wheels in the vertical axles.

5 Solid State Phenomena Vols Chassis level control. The task of this layer is to maintain the desired level of the platform. It is obvious that the level is not as important as the balance. It is less critical to drive for a while with a wrong level than to exceed some threshold value and fall. The robot is expected to move slightly tiled on the curvatures and while accelerating or decelerating. The control will work all the time adjusting the speeds of wheels to achieve desired level based on measurements from inclination sensors. The customized commands will be sent to balance control, from where they can be resent to motor controllers or just denied if balance is in danger, i.e., a reflex action is currently occurring. Increased robustness cannot be achieved here by a simple validation of upcoming commands. Level control is nothing but compensation of the terrain roughness and thus additional future information about the terrain structure would be essential. Unfortunately, such information is not easy to obtain and will not be available in the first robot prototype. Driving control. This layer is responsible for converting upcoming global movement instructions into commands that can be sent to separate motor controllers. As was already mentioned, before arriving to controllers, the commands are firstly interpreted by lower layers where they can be customized or denied generating appropriate feedback. Also here, as it is the last layer of the lower level, configuration abilities are going to be introduced. It will be possible to adjust control system parameters or even temporary or permanently disable the lower layers. In order to allow developers of higher levels to focus on actual tasks, e.g. path planning, rather than on control commands, the control instructions have to be simple and independent from hardware. Appropriate interface solutions compatible with our previous designs will be used. Each kind of motion will have a mode of operation assigned to it. The first mode is following a straight line in the desired direction. It requires just two parameters, i.e., the angle of movement and the speed. The angle can be recalculated by the layer into angles of particular wheels. The next mode is the turning in place, i.e., a central point rotation. Only the speed has to be specified as the central point of the robot is assumed to be known. Going further, parameters can be added to introduce a different point of rotation and it will be the next operating mode. It is easily possible to derive the mode that allows setting different angles and/or speeds for particular wheels but it should be avoided as it would require passing a lot of knowledge about the system from this layer into the higher level and would make the latter hardware dependent. The mode can be derived but should be used only for testing purposes. High level control system This level will be held on an additional onboard computer connected to the CAN bus. Two main tasks will be performed here: robot control and data logging. The latter will allow storing the robot state variables and demanded commands in non-volatile memory. This will allow further analysis and simplify debugging. In future, after building next prototypes, besides controlling a single robot this level might also control a team of robots to act in some well-defined mission. The communication between robots might be easily established by adding wireless interfaces to the computers. This level is highly dependent on the type of additional hardware. Without it the robot will only be able to move in a known environment having a fixed map and information from odometry. Supporting the robot with additional compass helps in determining the correct direction but still odometry is very prone to errors. A localization technique is essential to cancel the errors on some time intervals. Again, some additional hardware like proximity sensors or cameras is required. Then using appropriate techniques the robot position can be estimated more accurately, which increases the robustness. Using the same hardware and the mapping techniques, it is also possible to build the environment map given the robot position. Going further, it is even possible to perform both environment mapping and self localization simultaneously. This method is called SLAM and stays for simultaneous localization and mapping [5]. If robots cooperate in a team, it will be possible in

6 48 Mechatronic Systems and Materials III future to exchange environmental information between them, which may help in fast and robust exploration of unknown terrain. In order to develop an all-terrain vehicle, the high level will be supported by the SLAM technique. In the future, for a better support of this technique, also the inclination of the robot will be taken into account. Combining the information with readings from GPS, it will be possible to build high resolution 3D environmental maps that will include all the unevenness the robot will encounter on its path. Summary A concept of a structured control system architecture for a new all-terrain mobile robot was discussed. Also the original robot structure was shortly explained. Emphasis was put on the lower level control as it is the first step for building the prototype of the robot and failing here will have great impact on further work. The lower level was also subdivided into modules (layers) in the way that each of them corresponds to the task that can be considered separately. Those tasks are for controlling the motors, robot balance, level and finally steering the vehicle. The modulus structure is compatible with a software engineering approach and is of even greater important for robotics, where structure, hardware or requirements are very likely to change. Providing compatible interfaces, the modules can be exchanged with different solutions, which is very important for research projects. Layering was shown to have yet another important advantage, i.e., that it forces the message flow between the layers. That allows us to customize or even to deny the commands upcoming from higher layers. This way robustness can be increased by using appropriate models in the layers to predict the results of the commands. Future work will focus on synthesizing appropriate models and on increasing the robustness of the system. Acknowledgement This work is supported by the BMBF (Bundesministerium für Bildung und Forschung) ministry in the scope of the program IngenieurNachwuchs 40M. We thank the companies Daum&Partner and Buck Engineering and the university workshops for the kind support. References [1] G. Dudek, M. Jenkin: Computational Principles of Mobile Robotics, Cambridge University Press, Cambridge, 2004 [2] U. Nehmzow: Mobile Robotics: A Practical Introduction, Springer, Berlin, 2004 [3] M. Zając, R. Stetter, A. Paczynski: Concept for a Sensor/actor-network for the Robust Control of a Mobile Robot with All Wheel Steering, VDI-Berichte , num 1971, p [4] M. Zając, R. Stetter, A. Paczynski, D. Uciński: Concept of control system for an innovative mobile robot chassis, Proc. of 16 th International Conference on Control Systems and Computer Science - CSCS-16. Bucharest, Romania, 2007, p [5] R. Siegwart, I. R. Nourbakhsh: Introduction to Autonomous Mobile Robots, MIT Press, 2004 [6] G. Dudek and M. Jenkin: A Multi-Layer Distributed Development Environment for Mobile Robotics, Proceedings of the Conference on Intelligent Autonomous Systems (IAS-3), Pittsburgh, PA: IOS Press (1993) [7] M. Nikolaou: Model Predictive Controllers: A Critical Synthesis of Theory and Industrial Needs, Academic Press (2001)

7 Mechatronic Systems and Materials III / Control System for an All-Terrain Mobile Robot /